November 25-27

Adopting AI agents is one thing. Building a team of them that actually runs an organization’s work is another. Once an organization makes the leap from conversational AI to autonomous agents, the next question arrives quickly: how does a team move from a single agent doing one task to a coordinated system that handles entire functions of a business, end-to-end?
This hands-on session addresses that question. Drawing on years of building APG Media as an AI-first agency – with a library of custom skills, Model Context Protocol (MCP) servers connecting agents to live tools, and orchestration loops running across more than 70 clients – Aurimas Paulius Girčys opens up the actual machinery behind a working multi-agent operation. Attendees will see how a single instruction becomes a plan, how that plan is decomposed and delegated to specialized sub-agents, each with its own context, tools, and a narrow job, and where humans stay in the loop and why.
This is not a product demo. It is the architecture, the patterns, and the hard-won lessons of someone who builds these systems for a living – including what breaks, what is worth automating, and what absolutely is not. Aurimas Paulius Girčys covers skills as reusable building blocks, why a skill library, not a user interface, is the real asset, what MCP actually does to give agents “hands,” the orchestration pattern that separates planning from execution, and how to place approval checkpoints so that teams gain speed without losing control or accountability.
For technical operators, founders, and builders who have already adopted agents and now want to architect a coordinated system rather than a pile of one-off automations, this session is the blueprint for going from a clever automation to infrastructure that a business can run on.

The decisions that look free today can become expensive in six months – and AI accelerates the process at exactly the moment when the bill is quietly written. This talk is about that trade-off. The friction that vibe coding removes used to be where feasibility and constraints were discussed, so implicit decisions are made by the model instead.
In this talk, Alex Ponomarev covers where “free now” turns into “costly later”: blurred guardrails, code shipped without understanding its structure, and verification deferred. He shows how to price that deferred bill up front – by choosing which decisions to log and which to keep reversible. AI does not eliminate technical debt. It issues it faster and bills it on a delay.

Most organizations have given their people artificial intelligence (AI) tools and called it a strategy. The licenses are paid for. The output is generic. The transformation never came. The missing piece is not a better tool. It is organizational voice – the ability to encode who an organization is, how it thinks, and what it stands for directly into every AI assistant its team uses daily.

Stefan Chiriacescu built that system from scratch. The result: an 11× revenue increase, a team that moved from execution to strategy, and a 30-person organization operating with the output of an organization three times its size.

In this session, Stefan Chiriacescu will show attendees the live architecture behind that outcome – how to give AI systems a real organizational identity, how to map intelligence across every department, and how to turn every employee into the most capable version of themselves.

Not a concept. Not a demo. A working system attendees can replicate by Monday.

System landscapes are becoming increasingly complex. Teams place layer upon layer, abstracting away and hiding complexity. However, this creates new complexities that put engineering teams under significant strain and consume their mental capacity.

AI does not fix this because it has to run somewhere. And it comes with its own complex ecosystem of tools needed to bring AI to life.

In this session, Max Körbächer will introduce Naira, an open-source AI Engineering Hub that connects models, datasets, inference endpoints, MCP integrations, AI gateways, observability, documentation, and plugins across existing tools, so teams can understand, govern, and operate their AI landscape. Funded by the EU IPCEI-CIS initiative, it seamlessly integrates into Europe’s initiatives for digital sovereignty, AI empowerment, and strengthening transparency in AI usage.

Through an extensive plugin system, Naira is designed to connect with anything needed to build an AI-serving, tuning, and operating platform. It aims to reduce AI engineers’ cognitive load, simplify workflows by reducing tool jumps, and provide transparency across an ever-growing landscape.

Klemensas Mecejus will present a practical case study session on why Europe’s next AI race is not about model announcements, but about production workflows. Drawing on AI71 deployment work across government, construction, defense, telecom, and regulated enterprise sectors, the session explains how leaders move from pilots to governed, secure, and measurable AI operations. It covers workflow ownership, data access, cybersecurity, privacy, model risk approval, adoption, value capture, and the operating model required to industrialize sovereign and enterprise AI. Attendees will leave with a deployment checklist for selecting high-value workflows, preparing data and controls, aligning business and technical owners, and measuring whether AI is creating operational value rather than just producing demos.

AI prototypes are easy to demo and surprisingly hard to trust. A copilot gives a convincing answer, an AI agent completes a workflow once, and a prompt looks better after a few tweaks – but how can a team know whether the system actually improved? How can it catch regressions, measure quality, control cost, and decide whether a new version is ready for production?

In this practical session, Eugene Sergueev shows a step-by-step evaluation playbook for large language model (LLM) applications and AI agents. The talk focuses on moving from demo-driven AI development to a repeatable engineering process with real-world cases, quality rubrics, baselines, evaluation runs, thresholds, release gates, and production feedback loops.

The session walks through a lightweight reference architecture that teams can adopt without building a large evaluation platform first. Eugene explains how to collect 20–50 representative real-world cases, define what “good” and “bad” mean, create a golden dataset, compare prompt, model, and agent versions, and detect regressions before they reach users.

The talk also covers where rule-based checks, human review, and LLM-as-a-judge approaches work well – and where they fail. Eugene discusses how teams can use shadow runs, A/B tests, expert corrections, and production feedback to improve AI systems in small, measurable steps.

Attendees will leave with a practical framework they can reuse in their own teams: an evaluation loop, a quality rubric structure, a release-readiness checklist, and a reference architecture for evaluating LLM-powered workflows and AI agents.

The session is designed for engineers, AI practitioners, product engineering leaders, and platform teams who have already seen impressive AI demos and now need a way to make AI systems reliable, measurable, and safe enough for real production use.

Additional relevant information:

This is not a vendor or platform demo. The session focuses on practical engineering patterns that can be applied across different AI stacks: internal copilots, LLM-powered product features, document question-answering (QA) systems, AI-assisted development workflows, and tool-using agents.

The talk is structured as a step-by-step implementation guide and includes:

a lightweight evaluation loop for LLM apps and agents;
an example structure for a golden dataset;
a quality rubric for AI outputs;
examples of metrics and thresholds;
a release-readiness checklist;
common anti-patterns when teams evaluate AI manually;
guidance on where human review, automated checks, and LLM-based graders should be used.

Complex agentic systems can often chain dozens, or even hundreds, of sequential tool calls to complete a task. Near-instant resume performance is therefore a critical requirement for agentic systems running across multiple regions.

Traditional containers provide speed but not enough isolation. Full virtual machines (VMs) provide isolation but are too slow for ephemeral, per-task execution. When an agent needs to use a sandbox to execute code or access a filesystem, waiting seconds for a container or full VM boot can damage the user experience.

This session dives into the systems engineering behind Blaxel’s “perpetual sandboxes,” isolated compute environments that start and stop almost instantly and maintain state between shutdowns. Vikram Vaswani will explore how building an “agent-native” control plane using Kubernetes, Firecracker, and Virtual Kubelet allows for sub-25 ms resume times.

This architecture lets Blaxel sandboxes start in under 25 ms and remain on standby when not in use, eliminating cold starts without complex orchestration. It also provides airtight isolation and secure compute environments for agents to autonomously and safely run arbitrary commands and code.

Key Takeaways

* Understand how to build agent-native infrastructure on Kubernetes.
* Understand why cold starts severely degrade agentic user experience (UX) and throughput.
* Learn how sandboxes combine VM-level isolation with container-like responsiveness.

Audience

* AI application developers building or integrating high-frequency agents on Kubernetes.
* Architects and site reliability engineers (SREs) interested in new patterns that balance performance, security, and operational simplicity at scale.
* Platform and DevOps engineers responsible for building Kubernetes platforms with isolation, startup latency, and multi-region execution.

We are entering a new era in which an increasing number of individuals rely on AI chatbots and agents for information, advice, and even emotional support. Furthermore, there is growing discourse around the leaders of AI companies aiming to win the race to create an omniscient, omnipotent, and omnipresent AI “god” that could influence governance and politics.
This raises the question of what the role of AI professionals is in this new era, where old gods are being replaced by technological ones. The goal of this presentation is to discuss the most recent data on AI anxiety, AI ambivalence, and related psychological challenges associated with the adoption of AI systems. Povilas Godliauskas will also provide psychological tools and techniques to help attendees maintain purpose and mental health in the face of these new psychological and societal challenges.

While many organizations have successfully launched AI pilots, only a few have managed to transform artificial intelligence into a scalable business capability that delivers measurable impact across markets, teams, and clients.
In this session, Karin Immenroth, Global Chief Data Officer at Mediaplus Group, shares how one of the world’s largest independent media agencies built a global AI Operating Model that connects data, analytics, AI applications, automation, and business processes into a single ecosystem.
Drawing on real-world experience from the Mediaplus transformation journey, Karin will demonstrate how the organization established a global data mesh architecture, implemented a centralized data platform, deployed AI-powered decision intelligence solutions, and embedded AI into everyday workflows across international markets.
The session provides a practical blueprint for organizations looking to move beyond fragmented AI initiatives and scale AI across the enterprise. Attendees will gain insights into technology architecture, governance, operating models, organizational change, and adoption strategies that enable sustainable AI transformation.
Rather than focusing on theory, the presentation highlights concrete implementation decisions, lessons learned, measurable business outcomes, and practical frameworks that participants can apply within their own organizations.
Key Takeaways
Participants will learn:
How to move from isolated AI experiments to enterprise-wide AI adoption
How data mesh principles can accelerate scalable AI deployment
How to connect data platforms, AI applications, and business processes into a unified operating model
How agentic AI can automate decision-making and operational workflows
How to drive organizational adoption and build an AI-first culture
How to measure AI success through business impact, productivity gains, and speed-to-decision metrics

AI is here, and it is here to stay. So how can organizations avoid the fabled “dead internet,” where most content is generated by AI and consumed by AI, cutting the human element out of the system altogether?

In this talk, Darren Richardson explores the sociology of the internet, the effects of social media on media manipulation, the possibility of automatically generated content, and how solid DevOps tooling can be used to generate a set of guardrails for keeping AI on the path people want.